OPTIMIZATION OF COMPLEX HYDROLOGIC MODELS USING RANDOM SEARCH METHODS.

Abstract

The primary objective of automated calibration of complex conceptual hydrologic simulation models is to find the global optimum of a specified response surface. While direct search techniques such as gradient or Newton methods may be valuable tools for determining local optimum points, they present many practical and theoretical difficulties in real applications and often are of limited utility for global problems. As an alternative, four random search techniques have been proposed and analyzed in this study. A comparison experiment was performed on synthetic data using a state-space version of the Sacramento Soil Moisture Accounting Model. Experiment results are presented and implementation details are discussed.